Nvidia unveils powerful 'superchip' for AI

Nvidia has unveiled its latest chip for powering artificial intelligence [AI], known as Blackwell, as it seeks to consolidate its position as the major supplier to the AI frenzy.

"We need bigger GPUs. So, ladies and gentlemen, I would like to introduce you to a very, very big GPU," said CEO Jensen Huang at a developers' conference in California on Monday, referring to the graphics processors that are vitally important to creating generative AI.

The new chip has 208 billion transistors, more than double the 80 billion on the company's previous chip. All of those transistors can access the memory attached to the chip at nearly the same time, improving productivity.

"The launch of 'Blackwell' means Nvidia will capture even more capex dollars than its previous 'Hopper' generation GPU. We expect Nvidia will continue to not only lead but will widen the gap to its competitors in AI," Kinngai Chan, managing director and senior research analyst at Summit Insights, said after the announcement.

Nvidia said major customers, including Amazon.com, Alphabet's Google, Meta Platforms, Microsoft, OpenAI and Tesla, are expected to use the new chip.

Though Nvidia is widely regarded as a chip designer, the company has built a significant battery of software products as well.

Nvidia said earlier this year its software and services business had reached an annual run rate of $1 billion by the end of the company's last fiscal year.

'Insane' advance

Blackwell GPUs would deliver AI "superchips" four times as fast as the previous generation when training AI models, Nvidia said.

"The rate at which computing is advancing is insane," Huang said.

They would also deliver 25 times the energy efficiency, Nvidia said, a key claim when the creation of AI is criticised for its ravenous needs for energy and natural resources when compared to more conventional computing.

Initially developed to improve the graphics quality of video games, the company run by Huang figured out GPUs were perfectly suited for developing the large language models [LLMs] that underpin generative AI interfaces such as ChatGPT.

Unlike its rivals Intel, Micron and Texas Instruments, Nvidia, like AMD, does not manufacture its own semiconductors, but uses subcontractors, mainly the Taiwan Semiconductor Manufacturing Co.

Given the geopolitical concerns with Taiwan and China, this could be a potential weak spot, and the US has banned Nvidia from sending its most powerful chips to Chinese companies.